report error when using wildcard for top-level type parameter name

[scabug]

I am able to crash the compiler with the following code:

object TestModel {
}

case class CaseTest[_](someData:String)

The error is the following:

 java.lang.AssertionError: assertion failed: transformCaseApply: name = _! tree = _ / class scala.reflect.generic.Trees$Ident
at scala.tools.nsc.typechecker.RefChecks$RefCheckTransformer$$anonfun$2.apply$mcV$sp(RefChecks.scala:1493)
at scala.tools.nsc.typechecker.RefChecks$RefCheckTransformer.transformCaseApply(RefChecks.scala:1309)
etc..

(I really hope I was supposed to issue this as a bug, first time ever I do this kind of thing. If I shouldnt, I am very sorry for the inconvenience)

[scabug]

Imported From: https://issues.scala-lang.org/browse/SI-5606?orig=1
Reporter: Jonas Ohlsson (whiteancient)
Affected Versions: 2.9.1

[scabug]

@adriaanm said:
thanks for reporting! yes, that should give a clean error instead of crashing

[scabug]

@adriaanm said:
this should give a parse error

[scabug]

@paulp said:
Interesting what happens when one tries to disallow it. (It's clearly a bug it was allowed - try using a second "wildcard" and it fails.)

[scalacfork] /scratch/trunk1/src/library/scala/collection/immutable/RedBlackTree.scala:88: error: identifier expected but '_' found.
[scalacfork]   def keysIterator[A, _](tree: Tree[A, _]): Iterator[A] = new KeysIterator(tree)
[scalacfork]                       ^
[scalacfork] /scratch/trunk1/src/library/scala/collection/immutable/RedBlackTree.scala:89: error: ']' expected but ';' found.
[scalacfork]   def valuesIterator[_, B](tree: Tree[_, B]): Iterator[B] = new ValuesIterator(tree)
[scalacfork] ^
[scalacfork] /scratch/trunk1/src/library/scala/collection/immutable/RedBlackTree.scala:89: error: identifier expected but '_' found.
[scalacfork]   def valuesIterator[_, B](tree: Tree[_, B]): Iterator[B] = new ValuesIterator(tree)
[scalacfork]                      ^

[scabug]

@som-snytt said:
This is degenerate, but ought it to be illegal?

In the real world, there are many behaviors that are degenerate but not illegal, or no longer illegal, except in certain states of the Union where they frown on that sort of thing. (For instance, buying beer on Sunday.)

I took a quick look, and the example CaseTest[_] fails in the "copy" method but not in the companion module factory method, which are not exactly the same for some reason. The error is the wildcard in the new expr. (-Xprint:typer shows new pkg.this.CaseTest[_](data) versus new pkg.CaseTest[Any](data).) So this comes close to passing under the radar.

The spec says that the motivation for the shorthand A[_] is visibility and relevance.

One use case for top-level _, in quick test code, is to stub some UserKlass[This, That, Other] as class UserKlass[_,_,_].

I have to come up with symbols? Couldn't the compiler just make me a symbol? So I don't have to remember if we prefer A, B, C or T, U, V? This goes to relevance. And it depletes my decision-making budget.

Another use case is where a type param is deprecated because improvements in type inference make it superfluous.

trait T[@deprecated("I'm free","2.11") _, B <: S[_]]

Then, hypothetically, one could warn on usage of T[X, Y] but not T[_, Y].

Just thinking aloud.

[scabug]

@paulp said:
It could be allowed to work; my point is that it would be better for it not to work at all than for it to accidentally work for the one-type-parameter case only.

[scabug]

@retronym said:
The failure mode changed in 2.10.0. In scala/scala#826, the return type of the synthetic copy method was inferred, rather than explicitly provided. If the type parameter is named _, it is inferred as CaseClass[Any].

~/code/scala2 scala210
Welcome to Scala version 2.10.0-20121024-085118-2c554249fd (Java HotSpot(TM) 64-Bit Server VM, Java 1.6.0_27).
Type in expressions to have them evaluated.
Type :help for more information.

scala> case class CaseTest[_](someData:String)
defined class CaseTest

scala> new CaseTest[Int]("").copy()
res0: CaseTest[Any] = CaseTest()

scala> case class CaseTest[A](someData:String)
defined class CaseTest

scala> new CaseTest[Int]("").copy()
res1: CaseTest[Nothing] = CaseTest()

scala> case class CaseTest[_](someData:String)
defined class CaseTest

scala> new CaseTest[Int]("").copy[Int]()
res2: CaseTest[Any] = CaseTest()

scala> case class CaseTest[A](someData:String)
defined class CaseTest

scala> new CaseTest[Int]("").copy[Int]()
res3: CaseTest[Int] = CaseTest()

scala> case class CaseTest[_](someData:String)
defined class CaseTest

scala> val x = CaseTest[Int]("")
x: CaseTest[Any] = CaseTest()

scala> x.copy
                                               def copy[_](someData: String): CaseTest[Any]  

I figured I'd be able to trip up Value Classes with this, but while we can generate some funny looking signatures, they seem to go unnoticed.

scala> class CCC[_](val value: String) extends AnyVal { def foo[_]: Option[`_`] = None }
...
          final def foo$extension[_ >: Nothing <: Any, _ >: Nothing <: Any]($this: CCC[_]): Option[Any] = scala.None;

That sort of signature comes up without the _, of course, but seems to be innocuous as everything is symful by that stage.

scala> class C[A](val value: String) extends AnyVal { def foo[A] = this }
...
          final def foo$extension[A >: Nothing <: Any, A >: Nothing <: Any]($this: C[A]): C[A] = $this;

[diesalbla]

@SethTisue Since the code in the original example already compiles (so there is no longer a compiler crash) and there already is a pos test case in the compiler, should this still be an open bug?

[som-snytt]

It doesn't crash, but retronym's copy example is still wrong. Also C[_,_].